Volume 7, Issue 3, May - June 2018 Design of an Embedded ... · module and Raspberry Pi. [2] Tony Di Cola: This work “Raspberry Pi Face Recognition in Treasure Box” is a great

International Journal of Emerging Trends & Technology in Computer Science (IJETTCS) Web Site: www.ijettcs.org Email: [email protected], [email protected]

Volume 7, Issue 3, May - June 2018 ISSN 2278-6856

Volume 7, Issue 3, May – June 2018 Page 18

Abstract:Nowadays, identity thefts and frauds have become a major issue in our society. To mitigate the fraudulent activities, it is necessary to make use of a sophisticated facial recognition system. The objective of the project is to build a surveillance system based on facial detection and recognition technologies. This paper presents details on a facial detection and recognition algorithms and their implementation on Raspberry Pi device. The algorithms performed by Open CV library functions were modified for effective operation on the mentioned platform to use it as an embedded surveillance system. To achieve greater accuracy and efficiency, Open CV libraries with C++, Python computer languages were used. Training and identification were performed on an integrated device - Raspberry Pi3 B. We have created an advanced surveillance camera capable of detecting faces and, at the same time, recognizing the detected face. Face detection and facial recognition were performed with the Open CV library. All the processing was carried out on Raspbian OS on Raspberry Pi 3 B device. To capture pictures, we use camera module. This ability of facial recognition can help us to improve security systems. Keywords:Face recognition, Face detection, Raspberry Pi 3 B, Open CV, Video surveillance. 1. Introduction Now a days, many incidents such as robbery, stealing and unwanted entrances occur unexpectedly. So security is needed in our daily life. People remain busy in their day to day work but always would like to ensure safety of their beloved ones. Sometimes, people may forget to keep track of necessary items such as wallets, keys etc. Without these, they can’t enter home or other places they want. Traditional security system requires a key, a security password, an RFID card, or ID card for access. However, these security systems have shortcomings; for example, unauthorized people can steal the key or card leading to hazardous entry. With this in mind, a software has to be developed that can guarantee a higher level of security in a model. One of the unique features of our brain is that it can think only in images not in words. Each one of us has a unique face and it is the most important part of our body. From many years, we are using smart cards, tickets, PINS, RFID, keys etc., for authentication and to get rightful access to confidential areas such as ISRO, NASA, and DRDO etc. There are two types of biometrics designed till date - physiological characteristics (face, fingerprint, finger geometry, hand geometry, palm, iris, ear and voice) and behavioral characteristics. Sometimes your behavioral traits may

change because of illness, fear, hunger etc. Facial detection and recognition system is cheaper, simpler and does not cause disruption. The system has two processes - face detection (1:1) and recognition of the detected face (1: N). In face detection we classify between face regions versus non-face regions where as in the recognition process we compare the single face image detected with multiple images from input images. This project work uses BCM2837 processor, popularly known as Raspberry pi model. The core of the board is the above processor. It is a RISC processor based on ARM11. Surveillance systems with video plays a major role in many fields such as banking, personal security, business etc. Video surveillance has become important for everyone starting from small houses to huge industries and it helps fulfilling our safety aspects in many ways. Surveillance means monitoring any situation from a distance using cameras, binoculars, etc. This process is very much necessary to enforce law by investigating or preventing criminal activities and recognizing threats. We will build an embedded surveillance system with Raspberry Pi 3B device. Major portion of this work is built on study of selected functions used in latest surveillance systems such as the advanced algorithms to recognize human faces. After selecting appropriate algorithm, we develop the same with the help of C ++ and python programming language to use the computational power of embedded minicomputer to the full extent. We have a camera that captures still images and these images goes in as an input for image processing algorithms [1]. These algorithms processes images in real time to generate required information. To automate these activities, the basic approaches of Computer Vision are developed further and then applied to a real-time camera feed. The functions are present in the Open CV library (Open Source Computer Vision), an open source library with many algorithms to analyze and to manipulate images and video. It is totally compatible with most of the operating systems such as Windows, Linux, Android and Mac OS. It also has MATLAB interface and C ++, C, Python, Java integration. The functions are fully developed using Open CV libraries and necessary optimizations were carried out for effective functioning on Raspberry Pi platform.

Design of an Embedded Surveillance System using Raspberry Pi

Pavan Reddy Punnam1*,Dr.Munaswamy Pidugu2

1Student, Department of Electronics and Communication Engineering, Institute of Aeronautical Engineering, Hyderabad,

India. *Corresponding author

2Professor and Dean, Department of Electronics and Communication Engineering, Institute of Aeronautical Engineering,

Hyderabad, India.

mailto:[email protected]




2. Literature Review Raspberry Pi models were widely used in digital image processing area. We have many papers illustrating techniques such as image capture with embedded system on Pi, speaker and facial recognition, password key etc. The facial recognition systems are designed and improved almost daily for security reasons and need. Various parameters such as rejection rate due to false positives, accuracy of detection are calculated as an indicator of performance over using non-living things such as smart cards, physical keys for access. Hence various parameters such as performance, speed, are very important while choosing hardware design of Raspberry Pi models. Other unique features are versatility and low cost with respect to display modules. As days pass by, people will utilize Raspberry Pi as the main module to fetch necessary information. [1] K.Gopalakrishnan, V. Sathish Kumar: He used a unique and easy to implement embedded platform. The design was built on image capturing technique based on Raspberry Pi board. Based on the requirements of image recognition algorithm, Raspberry Pi processing module was implemented for this platform. Finally, it was concluded that the designed embedded platform is quick in processing recognition algorithm after capturing images and making sure that the data stream flow is smooth between camera module and Raspberry Pi. [2] Tony Di Cola: This work “Raspberry Pi Face Recognition in Treasure Box” is a great example on how Raspberry Pi and Pi Camera with Open CV’s algorithms can be put to real time use. By using the latest version of Open CV software, the device can access latest and most interesting algorithms on facial recognition. He also implemented Solenoid double action lock to use lock the device after power off. [3] Kuldeep Soni: He developed an advanced surveillance system that can detect and recognize a face at the same time. He used Open CV libraries on Raspbian OS. To capture images, Pi Camera Board was used. With the help of recognition capability along with detection he was able to prove that this is a highly secured system [4] MedakTeena Ravali, Prof. Ranga Sai Komaragiri: In their work, Raspberry Pi board with Open CV packages was proposed as an alternative to DSP kits for image processing. In this work, the algorithm for facial recognition was designed on PCA (Principal Component Analysis). This system was designed on criteria such as resource optimization, less power consumption and high operational speed. [5] Anoop Mishra, Dixit Arshita: Their work was on Raspberry Pi 2B+ model with camera interface. It then converts the captured image in to gray image with digital processing algorithm. Also, they concluded that the results are rational and can be practically applied. This device is

technically smarted than performing image processing on a personal computer [6] KandlaArora: Real time application of facial recognition was designed on MATLAB. The approach was based on PCA with Eigen faces. [7] AjinkyaPatil, MrudangShukla: They built a facial recognition system to use it for student attendance in their class. This helped them save time as it is an automated process based on image processing. The device can detect and differentiate faces from non-faces to get an accurate attendance. The backend database has student names, images of student’s face and their roll numbers [8] Manal Abdullah, MajdaWazzan and SaharBo-saeed: Their work is based on optimizing time complexity of PCA. They proposed an enhanced algorithm with PCA that does not affect the performance of recognition system. [9] Official Raspberry site - http://www.raspberrypi.org 3. 3. Hardware and Software Used 3.1 Raspberry Pi To identify the right hardware is a vital first step for this project work. We chose a Raspberry Pi 3B model after doing a lot of research and comparing micro controllers on various aspects such as cost, processing speed and user friendliness. Unique features such as high processing capacity, relatively lower price and adaptability to different programming modes stands out to be the main reasons to choose this model. The device has Linux operating system and has access to a lot of libraries and compatible applications. Raspberry Pi has an Ethernet port for network connection and can be managed as long as we are in the same subnet. It also has four USB sockets and an HDMI port that will help to look at the interface of operating system. It has around 40 pins to receive and send signals. These pins can be divided in half in to 2 categories – 3V and 5V groups. With this bifurcation, one side of the micro controller has a voltage of 3V and the other side has 5V. There is no operation system installed in a new Raspberry Pi. We have to download the operating system from official Raspberry website and transfer it to a SD card. Fig.1 depicts Raspberry Pi model B3 and its components. The main foundation has Debian, Arch Linux ARM distributions and uses Python as the main programming language, with the support for BBC BASIC [5], and Perl. We developed algorithms, for face detection and recognition using Python language.


http://www.raspberrypi.org




Figure 1Raspberry Pi 3B Hardware

3.2 Raspberry Pi OS To install the operating system in Raspberry Pi, we have to download New Out of the Box Software – NOOBS. It will help us set up Raspberry Pi. Immediately after we boot up NOOBS, we get a selection of operating systems to choose from and this list depends on the Raspberry Pi model number.

Figure2List of OS Window

Raspbian is the ‘official’ operating system of Raspberry Pi device. It is similar to Linux operating system that was built for Raspberry Pi. You will see all software required to carry out basic tasks. It also has Office suite by Libre, an email program to send and receive emails, a browser to access web and other few applications to learn how to program.

Figure 3 Raspberry Pi Home Window

4. Methodology

Figure4 Block Diagram

Figure5 Required Equipmentfor Project

Power Supply: The Pi needs a quality power supply (>= 2A at 5V for Model 3B or =700mA at 5V for old models). We can also use a 5V micro USB cable to supply power. Keep an eye on the power supply, as less power will make the Pi to behave differently. Camera Module: Camera module is LM (Linear Motor) camera with USB interfacing to the raspberry pi module. Its resolution is 25-megapixel and has 6 Flash Lights with maximum image transfer rate of 1080p: 30fps. Image gets transferred to Raspberry Pi module once the camera captures image. At first camera module captures 50 images to create a database of train faces of the authorized person. Secondly, it take a test face or live captured image to compare it with train faces. Ethernet:





Raspberry Pi is connected to a network or internet with a standard LAN cable on the Ethernet port [9]. Display Monitor: To help the user to see notifications from Pi HDMI cable: High definition multimedia interface (HDMI) port is provided on the board for connecting raspberry Pi to monitor. SD card: An 8GB class four SD card is used with NOOBS software. Keyboard and mouse: We can use a standard model for keyboard and a basic mouse 5. Face Detection Face Detection using Haar Cascades is the algorithm we will use for this project. In this algorithm the object based detection happens using cascade classifiers. Cascade classifiers are based on Haar features and this algorithm is an optimal object detection method as mentioned by Paul Viola, Jones Michael in their paper. In 2001, there was another machine learning based model where the cascade features are built with training on positive and negative images. This machine learning model was called as ‘Rapid Object Detection using a Boosted Cascade of Simple Features’. This machine learning technique, the model requires a lot of data points (picture of human faces and pictures without human faces) to learn from. Based on the features extracted from the data points provided, the model will understand the underlying patterns and will be able to distinguish a new image provided it. Features get designed by subtracting pixels count of white rectangle from pixel count of black rectangle. A lot of features can be derived by knowing all possible sizes and locations of every kernel. Calculation of pixels under white rectangle and that of black rectangles is a must for feature extraction. For this we use integral pictures to simplify calculation of sum of pixels. It doesn't matter how bulky the number of pixels are, it will compress them into just four pixels. This reduction in pixel size helps the algorithm to run very fast. Among all the features designed, most of them might be inappropriate as they won’t be able to explain the required variance. To identify features that are useful, we analyze each of the features on all images from database. By doing this, we will categorize faces to positive or negative for every feature. Apparently, there will be misclassifications while doing this exercise. We only go with the features that have less fault rate i.e., the features that can classify faces and non-faces properly. The actual procedure might not be as simple as this. Each feature is given equal weightage in the starting and after every iteration, weights will get changed (increased/decreased) based on the classification of pictures. We repeat this process to calculate new fault rates. We have to conduct multiple iterations until we reach necessary accuracy or a minimum allowable fault rate.

(a)

(b)

(c)

Figure 6 (a) Edges (b) Lines (c) Center Surround

Figure 7 Created database

6. Face Recognition We will work on real time application of Principal Element Inspection using Eigen faces. As soon as we have the face image, we compute the weight against the face images present in the back end database. The image that gets processed is linearly projected to a low dimensional image space and the difference with respect to a set of Eigen vectors is weighed. If the variance (weight) is below a certain verge, the image gets classified as a known face. Otherwise, the face gets tagged as an unknown face or not a face at all. We use Python code to start the webcam and capture images. Then the program matches the given image with set of images present in the back end system. For facial recognition, a similarity score gets calculated – similarity between the input image and every image in the database. The image or face with maximum similarity/nearness score shows assurance of the exact match. In the later sections of this paper, we will try and understand different stages in performing facial recognition. Now, we will focus on learning more about the Eigen faces.

Figure 8 Performing Face Recognition





A. Eigen Faces Explained: Eigen vectors get used in most of the compute vision problems to recognize human faces. In this paper, we get Eigen faces from the principal component analysis of distribution of faces. Basically, they are the Eigen vectors of a covariance matrix of images from training database, where an image is seen as a vector with NxN pixels in a N2-dimensional space. We use these principal components to categorize human faces. This technology was found and developed by Kirby, Sirovich in 1987. In 1991 Turk and Pentland implemented this for facial recognition. Eigen vectors are sorted accounting for different amount of variation. These set of vectors can explain the variation between face images. Each image location translate into an Eigen vector and we can display it as a “shadowy” image called Eigen face. B.Facial Recognition – Approach with Eigen Faces: Approach for facial recognition has below steps: 1. Initialization: We have 2 steps in this procedure: -Fetch and store as many training images as possible -Compute Eigen faces on training dataset keeping only the finest M images with maximum eigenvector values. These images define the “free space”. As we keep adding more faces, the Eigen vectors should be updated by allowing the model to learn from new faces -Analyze and calculate the weightage by projecting facial image on to training images in M-dimensional space 2. Recognition: After initialization, we start with recognition process: - To analyze and recognize an image, we first compute weights of M Eigen vectors by overlaying this image on to every Eigen face created from training database - If the image gets classified as a face, check other patterns to identify if it's a well-known person's face or an anonymous one - (Optional) although this is an optional step, it is helpful to update the Eigen faces and/or weight patterns learning from the new image - (Optional) analyze the weight patterns of the new image and add it to the list of known faces in the backend database.

Figure.9 Eigen Faces of Corresponding Training Images

Set.

7. Flow Chart

Figure 10 Flow Chart of Project

In above fig we can see the flow of various stages of project. Firstly, when we are done with the connections, turn on the power supply to initialize raspberry pi and camera. Secondly, when the authorized person comes in front of camera, the camera module will capture the face image with current postures. The captured face of current postures creates a data base of the authorized person and stores this. Then the camera module will capture the current live face of the person. All this process is done in Raspberry pi module. If a person comes in front of a camera the raspberry pi module will compare and check with the data base of the authorized person. When comparison is successful, the raspberry pi will capture an image of that person and sends an email along with image to an organization/ authorized person that person is “PERSON MATCHED”. If the comparison is unsuccessful, it will send a message that the person is “UNAUTHORIZED”.

8. Results

I performed a demo and found results of authorized person and unauthorized person.

Figure 11 Authorized Person’s Captured Image





Figure12 Unauthorized Person’s Captured Image

Figure 13 Setup of the Project

9. Conclusion The facial recognition system designed has applications in several sophisticated automated systems and public places such as hospitals, labs with high-tech equipment, banks etc. This system will help in reducing the hazard of unauthorized entry. As the system captures data of person faces, it will be useful in providing evidence to security department officials in the event of burglary. As the system is developed on Raspberry Pi, it is smaller in size, light-weighted and consumes very less energy. Hence, it is more easy to use than a PC-based facial recognition system. Also, we used open source code in designing the system making it almost as a free software developed on Linux. As we need the system to conduct real time face detection and face recognition, we used programming languages C++ and Python to carry out these operations efficiently. Face detection rate is the metric we used to determine the efficiency of the system. Extensive study revealed that the system designed displayed excellent performance in terms of face detection rate and the system can even to the same with poor quality images. Acknowledgment Thanks to Professor and DeanDr.MunaswamyPidugu, Electronics and Communication Engineering Department, Institute of Aeronautical Engineering, Hyderabad, India. References [1] Gabor Arva, Tomas Fryza. "Embedded video

processing on Raspberry Pi", 2017 27th International

Conference Radioelektronika (RADIOELEKTRONIKA), 2017.

[2] Shrutika V. Deshmukh, Prof Dr. U. A. Kshirsagar “Face Detection and Face Recognition Using Raspberry Pi” International Journal of Advanced Research in Computer and Communication EngineeringVol. 6, Issue 4, April 2017.

[3] Shrutika V. Deshmukh, Prof Dr. U. A. Kshirsagar “Implementation of Human Face Detection System for Door Security using Raspberry Pi” International Journal of Innovative Research in Electrical, Electronics, Instrumentation and Control Engineering ISO 3297:2007 Certified Vol. 5, Issue 4, April 2017

[4] Srinivasu Batchu, S. Praveen Kumar “Driver Drowsiness Detection to Reduce the Major Road Accidents in Automotive Vehicles” International Research Journal of Engineering and Technology (IRJET) Volume: 02 Issue: 01 April-2015.

[5] M. Narender, M. Vijayalakshmi. "Raspberry Pi Based advanced scheduled home automation system through E-mail", 2014 IEEE International Conference on Computational Intelligence and Computing Research, 2014.

[6] Kandla Arora “Real Time Application of Face Recognition Concept” International Journal of Soft Computing and Engineering (IJSCE) ISSN: 2231-2307, Volume-2, Issue-5, November 2012.

[7] B. C. Lovell, S. Chen, T. Shan. Real-time Face Detection and Classification for ICCTV.. [Online]. Available: http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.331.6676rank=1

[8] R. Szeliski. Computer Vision: Algorithms and Applications. [Online]. Available: http://szeliski.org/Book/

[9] slideshare [Online] Available www.slideshare.net [10] OpenCV. [Online].Available http://opencv.org/ [11] OpenCV: Face Detection using Haar Cascades

[Online].Available:https://docs.opencv.org/3.2.0/d7/d8b/tutorial_py_face_detection.html

AUTHORS

Dr.MunaswamyPiduguworking as a Professor and Dean of Electronics and Communication Engineering Department, Institute of Aeronautical Engineering, Hyderabad, Telangana, India. He has more than 18

years of teaching experience in various Engineering colleges. He completed his Ph.D. in Electronics and Communication Engineering from Jawaharlal Nehru Technological University, Hyderabad. He guided many PG and UG projects. He attended more than 3 international conferences and submitted 6 international journals. His areas of interest are Instrumentation and Control systems.

Pavan Reddy Punnam received his B.Tech. in Electrical and Electronics Engineering from Kakatiya University, Warangal. He is currently pursuing M.Tech.

in Embedded Systems, Department of Electronics and Communication Engineering, Institute of Aeronautical Engineering, Hyderabad. Control systems is his favorite area.


http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1

http://szeliski.org/Book/

http://www.slideshare.net

http://opencv.org/

https://docs.opencv.org/3.2.0/d7/d8

Volume 7, Issue 3, May - June 2018 Design of an Embedded ... · module and Raspberry Pi. [2] Tony Di Cola: This work “Raspberry Pi Face Recognition in Treasure Box” is a great

Documents